dual-review

Run sequential Codex adversarial reviews of specs and plans.

Updated Mar 26, 2026

Search Tags:#codex #structured-output #json-schema #plan-review #adversarial-review #spec-review

Ready to run

npx skills add https://github.com/mrkhachaturov/rkstack --skill dual-review-mrkhachaturov

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: dual-review
Source: https://github.com/mrkhachaturov/rkstack/tree/main/skills/dual-review
Command: npx skills add https://github.com/mrkhachaturov/rkstack --skill dual-review-mrkhachaturov

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Sequential Claude-Codex review loop that adversarially interrogates specs and plans to surface gaps, inconsistencies, and risk before implementation.

Core Features & Use Cases

Embedded in brainstorming and writing-plans to automatically run Codex reviews after self-review.
Standalone usage via the /dual-review command to audit any artifact with configurable rounds and an optional focus.
Returns structured JSON findings compliant with review-schema.json, enabling automated triage and fixes at the document level.
Iterative loop until approval or max rounds, with guidelines to edit the spec/plan rather than code.
Handles linked specs in plan mode and respects project conventions via CLAUDE.md and detection caches.

Quick Start

Invoke the skill on a target artifact using the /dual-review command to start the adversarial review loop.

Frequently Asked Questions about dual-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I run an adversarial review on a project plan?▼

Run an adversarial review on a project plan by invoking the /dual-review command, which orchestrates sequential Codex interrogations to surface gaps, conflicts, and risks. The loop iterates until approval or a maximum round limit is reached, outputting a structured JSON verdict.

How do I get structured JSON output from a spec review?▼

Structured JSON output from a spec review is generated by running sequential adversarial review loops that conform to review-schema.json. This process validates specifications and plans, producing a structured JSON verdict and recommendations that enable automated triage and traceable document edits.

Can I use adversarial review loops with brainstorming and writing-plans?▼

Yes, adversarial review loops can be embedded directly within brainstorming and writing-plans workflows to automatically run Codex reviews after self-review. This integration handles linked specs in plan mode and respects project conventions via detection caches.

Does the dual-review skill support configurable review rounds and focus text?▼

Yes, the dual-review skill supports standalone invocation with configurable review rounds and an optional focus text. This allows you to target specific areas of an artifact during the adversarial review loop until the spec or plan is approved.

What are the limitations of automated spec and plan reviews?▼

A key limitation of automated spec and plan reviews is that guidelines direct you to edit the documentation rather than the code. The review loop is constrained by a maximum round limit and focuses on surfacing risks in linked specs rather than generating code fixes.